Marine hydrodynamic bearings are mechanical components that guide ship shaftlines in rotation (submarines and surface ships). Lubricated with sea water, these crucial components in the propulsion kinematic chain are subject to very severe use that can disturb the operation of vessels.

The purpose of this thesis, which is part of the PEA INCOLA (INteractions COque Ligne d’Arbres - Shaftline Hull Interactions) financed by DGA, is to study the marine bearings using a numerical/experimental approach. A representative test bench (scale 1:10) designed for the occasion is used to characterise the operating conditions of a bearing (speed, temperature, misalignment, loading) and identify its critical phases.

These results are compared with predictions by a hybrid computer code developed for the occasion combining i) finite elements for the distortion of the entire shaft and ii) dry, lubricated or mixed contact depending on the conditions. The nominal conditions are studied as are the influence of wear and transient regimes.

Planning a naval mission tactically involves reconciling the choice of optimum route and handling of potential threats. This planning is re-assessed constantly as the mission unfolds.

The command analyses the information about the ship and its environment to (1) decide on new actions to be undertaken and (2) assess the success of the mission in real time. It is assisted in this by indicators and operation methods, including Recognition Primed Decision (RPD).

Although it is increasingly difficult to apply the RPD given the growing number of data to be included, the cognitive overload thus generated and time constraints for the operator, this method is still the most used for conducting naval missions. We propose under this thesis to formalise a decision-making aid environment based on the RPD to assist command during a mission.

The system will analyse in real time the dynamic evolution environment of the ship and propose actions to command that respond tactically to the mission requirements. The proposals are generated using deep reinforcement learning. The training incorporates diverse scenarios so that the system can adapt to totally new situations.

Lastly, a mapping interface is used to assess the proposals and make the appropriate tactical decision.

In terms of acoustic discretion, the periodic component of noise radiated by marine propellers is of particular concern, as it can be picked up at long range by towed passive arrays, which are particularly sensitive to ultra-low frequency noise.

The long wavelengths associated with these frequencies also cause complex interaction phenomena with the surface (Lloyd's mirror) and the hull, which require a dedicated theoretical analysis. Mounting propellers immediately next to the hull is in fact likely to amplify the radiated sound tremendously.

The propulsion system integration environment is thus crucial to the acoustic discretion performance.

The subject of this thesis is the generation of the periodic component of the noise radiated by the ship's propellers, its amplification by the ship's hull and the radiation of these acoustic waves through the water.

The main purpose of this presentation will be to reveal the effect of acoustic installation. This phenomenon is in particular pronounced for the tonal noise sources at ultra-low frequency, where an elementary analytical model suggests potential application of up to +20 dB.

Under the digital transformation and IoT (Internet of Things), new maintenance opportunities can be envisaged by constituting voluminous databases from sensors.

The PHM (Prognosis Health Management) methodology can be used in particular to move from systematic maintenance to conditional maintenance, to make replacements based on the remaining lifetime of different equipment. The maintenance policy can then be adapted to promote more dynamic logistics and through-life support.

In this thesis, we study the application of the PHM methodology to propulsion systems. We use especially deep learning methods to detect large-scale time series anomalies and the different architectures used to predict the remaining lifetime.

This type of model will subsequently be extended to constructing indicators and synthetic data available to maintenance operators to monitor the engine condition.

The interaction between a turbulent flow and a submerged system can product wide band acoustic radiation. This information is essential at all stages in the life of the system.

The acoustic radiation includes a vibro-acoustic contribution (excitation of the structure by the turbulent flow) and a purely hydrodynamic, so-called hydro-acoustic, contribution, governed specifically by Lighthill's equations. These describe the propagation of the radiated acoustic noise by the turbulence, taken as source term, in presence of the structure.

The purpose of this thesis work is to develop a new, semi-analytical predictive method of this hydro-acoustic component that can be applied to actual geometries with low Mach numbers and high Reynold's numbers. For this purpose, we use an integral formulation to bring the problem to the boundaries and a well-chosen Green's function that simplifies the problem further.

We also propose a new statistical turbulence model to approximate the source term of the Lighthill's equation and lastly we estimate the radiated acoustic noise using a stochastic squaring method.

The predictions have been validated in academic configurations and the model is being verified in realistic industrial geometries.